Ultrasound-guided IA injections of LPS (10 mg; from Escherichia coli 055; Sigma-Aldrich, Australia) were administered to one amniotic sac of twin-bearing ewes (Border Leicester; n = 12) at 118 ± 2 days of gestation (mean ± SD). Lambs were delivered by caesarean section 7 days later, dried, and a transcutaneous oxygen saturation probe was placed around a forelimb (SpO₂; Masimo, CA, USA). The lambs were intubated and then randomised to receive either a protective ventilation strategy (LPS + PROT, n = 6) or injurious ventilation strategy (LPS + INJ; n = 6), as described previously (5). Briefly, lambs in the LPS + INJ group underwent immediate cord clamping followed by 15 min of high V_T ventilation (targeting 12–15 mL/kg) using a neonatal positive-pressure ventilator (Babylog 8000+, Dräger, Lübeck, Germany). After 15 min, LPS + INJ lambs received surfactant (100 mg/kg, Curosurf^R, Chiesi Pharma, Italy), followed by tidal ventilation targeting V_T of 7 mL/kg. LPS + PROT lambs received prophylactic surfactant, a single 30-s sustained inflation at peak inspiration pressure (PIP) 35 cmH₂O (Neopuff Fisher & Paykel Healthcare, Auckland, New Zealand), followed by tidal ventilation (7 mL/kg) for 3 min prior to umbilical cord clamping (26). V_T targeted 7 mL/kg for the protective strategy for the duration of the experiment.

After delivery, umbilical arterial and venous catheters were inserted for measurement of blood pressure, arterial blood gas sampling, and invasive real-time monitoring of heart rate (HR) (Powerlab: ADInstruments, Castle Hill, Australia). Anaesthesia was maintained via intravenous infusion of Alfaxan (10 mg/kg/h in 5% glucose, Jurox, Auckland, New Zealand). Samples of arterial blood were collected every 5 min for the initial 15 min and at 15-min intervals thereafter for measurement of the partial pressure of arterial carbon dioxide (PaCO₂), oxygen (PaO₂), oxygen saturation (SaO₂), and pH (ABL30, Radiometer, Copenhagen, Denmark). Ventilation and oxygen delivery were altered to target PaCO₂ of 45–55 mmHg and an oxygen saturation of 90–95%. Ewes were killed humanely (sodium pentobarbitone 100 mg/kg i.v.) immediately after delivery of the lambs, while still under general anaesthesia. Lambs were killed humanely (sodium pentobarbitone 100 mg/kg i.v.) after the MRI.

The lambs were transferred to the MR scanner 1 h after delivery. The radiographers and team involved in supporting lambs during MRI acquisition were blinded to group assignment of each lamb. Lambs were scanned in a supine position, and ventilation was maintained using a BabyPAC portable and MR compatible ventilator (Pneupac, Smiths Medical, UK). HR and SpO₂ were continuously monitored to guide anaesthesia maintenance, and blood gas parameters were recorded at 15-min intervals to monitor lamb well-being and guide mechanical ventilation. Scans were performed on a 3T MR scanner (Siemens Skyra, Erlangen, Germany) at Monash Biomedical Imaging, Monash University (Clayton, Australia), with a 15-channel radio frequency (RF) coil for both RF transmission and reception. The MRI protocol comprised structural imaging sequences (T1, T2, and DTI), susceptibility-weighted imaging (SWI), and a single-voxel MRS (6). The total acquisition time was about 40 min.

A 3D MPRAGE sequence was applied to achieve T1-weighted images, 102.5 mm thick, for whole brain coverage. The repetition time (TR) was 1,440 ms, with echo time (TE) of 3.92 ms, an inversion time of 900 ms, and a flip angle of 9°. The data matrix was 256 × 256 × 128, collected from a field of view (FOV) of 200 mm × 200 mm, giving a voxel size of 0.78 mm × 0.78 mm × 0.8 mm.

T2-weighted images with a 3D SPACE sequence, 44 mm thick, were also obtained to provide whole brain coverage (TR = 1,000 ms, TE = 132 ms, flip angle = 120°, and averages = 2). The matrix size was 384 × 384 × 88, collected from a FOV of 192 mm × 192 mm, giving a voxel size = 0.5 mm × 0.5 mm × 0.5 mm.

A standard 3D gradient echo sequence was used to acquire SWI for the whole brain with a slab thickness of 72 mm (TR = 28 ms, TE = 20 ms, and flip angle = 15°). The matrix size was 288 × 216 × 60, collected from a FOV of 176 mm × 132 mm, giving voxel size = 0.6 mm × 0.6 mm × 1.2 mm. The second phase encoding direction for 1.2-mm voxel was acquired in a parallel direction to the main magnetic field to optimise the susceptibility effect.

Diffusion tensor imaging with a spin-echo echo planar imaging sequence in the axial plane was obtained with 50 contiguous 1.2-mm slices (TR = 11,400 ms, TE = 99 ms). The data matrix was 128 × 128, collected from a FOV of 154 mm × 154 mm, giving a voxel size of 1.2 mm × 1.2 mm × 1.2 mm; for five b0. Diffusion encoding gradients were acquired in 30 directions with a b value of 1,000 s/mm². The diffusion-weighted imaging was performed twice, and five b = 0 volumes were acquired.

A single-voxel spin-echo sequence was used to acquire MRS, localised on the supratentorial deep grey matter and central white matter (TR = 2,000 ms, TE = 270 ms) with a voxel size of 15 mm × 15 mm × 20 mm.

MRI examination, image processing and analysis, and interpretation of the structural MR images were performed completely blind. Once ROI was drawn, the group allocation of each lamb was revealed for the final statistical analysis. The colour mapping approach was then completed, but not performed in a blinded fashion. Images acquired from T1, T2, and SWI scans were used to determine the presence of gross brain injury including infarcts and haemorrhages. DTI images were converted into NIFTI format, and then a brain mask was manually placed over the image using the FMRIB Software Library [FSL, FMRIB, Oxford, UK (27)]. DTI data were corrected for eddy-current distortion and then DTI parametric maps: fractional anisotropy (FA), axial diffusivity (AD), radial diffusivity (RD), and mean diffusivity (MD) were created with the Diffusion Toolbox of FMRIB. AD was used as the largest eigenvalue of the tensor, and RD was used as the average of the second and third eigenvalues of the tensor. All maps were then coregistered with high-resolution T2 images of the brains using Linear Image Registration Tool [FLIRT (28)]. The b0 image was used as the reference for the coregistration.

ROI were defined manually on high-resolution T2 images and then exported to the FA, AD, MD, and RD maps to calculate the region mean values for each lamb. ROIs were placed in specific regions including the thalamus (Th), periventricular white matter (PVWM), internal capsule (IC), frontal white matter (FWM), and the cerebellum (CB) vermis, targeting the stalk and midline white matter (Figure 1). The Sheep Brain Atlas from Michigan State University (Brain Biodiversity Bank, National Science Foundation) was used to facilitate accurate identification of the anatomical structures. The mean value of the adjacent four to five slices that defined a given ROI volume was calculated for each of the ROIs to test between-animal variability in diffusivity for both groups. MATLAB R2013a (MathWorks, MA, USA) was then used to calculate the SD for the ROIs of each region of the lamb brains to test within-animal variability (heterogeneity in DTI).

MRS data were processed using TARQUIN software to implement spectral fitting in the time domain (6). Peak area ratios for lactate (Lac) relative to N-acetylaspartate (NAA) (total of NAA and N-acetylaspartylglutamate), creatine (Cr; total of phosphocreatine and free creatine), and choline (Cho; total of glycerophosphocholine and phosphocholine) were calculated. In addition, we calculated peak area ratios for NAA/Cho, NAA/Cr, and Cho/Cr.

DTI data from the ROIs appeared to show more heterogeneity in the injured lamb brains. To further examine this variability, we utilised a colour map threshold technique to determine where the low diffusivity voxels were distributed in the DTI volumes. ImageJ (version 1.48, NIH, USA) was used to create a histogram of the distribution of the MR image pixel intensities at each position. For the purposes of comparison between animals, the histograms from all lambs in the two groups were normalised. The total pixel intensity distribution (histograms) for each lamb brain was compared between the minimum and maximum intensities within the brain for 256 bins in all DTI images. Histograms of the diffusivities for the whole brains of individual lambs in the two groups exposed to LPS were determined (Figure 2). Histograms were also plotted for the whole brains in the two groups of lambs that were not exposed to LPS in our previous study (6), as shown in (Figure 2). The histograms demonstrate the frequency distribution of MR image voxel intensity levels at each position. In the histograms, the black box represents 75% confidence interval (CI) of all voxel intensities observed in the distribution in protective lamb brains. The lower limit of this box was the lowest 10% of observed intensities. We then applied a threshold to the mapped intensities in the overlay in both groups (width ~10%). The upper threshold limit (Figure 2, green dashed line) was manually set at an intensity where voxels were almost completely absent across the whole brain in protective ventilation groups of lambs. However, coloured voxels were still evident at the edges of the brain mask in protective lambs. We found that the upper threshold of the colour mapping was consistent with the 90th percentile of intensities in the protective lambs, with 3–5% overlap in observed pixel intensities between protective and injurious ventilation groups (Figure 2). This process permitted mapping and quantification of the number of voxels of low diffusivity for AD, RD, and MD colour map images, using ImageJ.

Physiological data were compared using a two-way repeated measure ANOVA and Holm–Sidak post hoc analyses to determine significant interactions (Sigmaplot, Systat Software Inc.). The LPS + INJ group was compared to the LPS + PROT group to assess between-group differences in DTI and MRS parameters. The mean and SEM of voxel intensities within the predefined ROIs were calculated, and independent t-tests were used to compare treatment groups with the significance level set at P < 0.05. Data from the ROIs are presented as mean (box 5-95% CI of mean) with maximum − minimum error bars. Further, we calculated the SD for the ROIs in each lamb’s brain to test within-animal variability and to test the heterogeneity of DTI diffusivities for differences between groups using MATLAB (MathWorks Inc., MA, USA) software.

LPS was successfully injected into the amniotic sac as confirmed by electrolyte analysis of amniotic fluid aspirated at the time of injection and the presence of thickened fetal membranes characteristic of this experimental intervention observed at delivery (12).

Mean gestational age (125 ± 2 days), body weights (LPS + PROT: 3.4 ± 0.6 kg; LPS + INJ: 3.6 ± 0.2 kg), and sex distributions of the lambs (LPS + PROT: 3 males and 3 females; LPS + INJ: 2 males and 4 females) were not different between groups.

Physiological parameters were recorded for the initial 60 min, prior to transfer of the lamb into the MRI scanner. Peak inflation pressure (Figure 3A) was significantly higher in the LPS + INJ group than the LPS + PROT group during the first 15 min (P < 0.001), resulting in a significantly higher V_T during the initial 15 min than the LPS + PROT group (P < 0.001; Figure 3B). However, PIP was lower in LPS + INJ lambs once protective ventilation was initiated, likely due to the delivery of surfactant at 15 min in these lambs. Consistent with this contention, the LPS + INJ group demonstrated significantly increased lung compliance at 20, 30, and 45 min of ventilation (P = 0.02, P = 0.04, and P = 0.05, respectively; Figure 3F). PaO₂, PaCO₂, and pH were not different between groups (P = 0.38, P = 0.7, and P = 0.9, respectively; Figures 3C–E) nor was FiO₂ (LPS + INJ; mean 0.41 ± 0.07 vs. LPS + PROT; mean 0.31 ± 0.06, P = 0.20). Oxygen saturation (SaO₂) did not differ significantly between groups after the first 5 min of ventilation (P = 0.5; Figure 3G).

Carotid blood flow was variable within groups and between groups, but overall there was no significant differences observed (P = 0.8; Figure 3H). Mean HR tended to be lower at all time points in the LPS + INJ compared to LPS + PROT group (P = 0.09; data not shown).

No overt cerebral injury was seen on T1, T2, or SWI images in either group. The MRS peak area ratios of Lac to other metabolites (Cr, Cho, and NAA) (Figure 4) as well as NAA/Cho, NAA/Cr, and Cho/Cr were not different between groups.

While mean AD, RD, and MD values were lower in the LPS + INJ lambs compared to the LPS + PROT lambs in many of the ROIs (Figure 5), no statistically significant differences were found. However, lambs in the LPS + INJ group showed significantly increased heterogeneity of diffusivity in multiple regions [Th, PVWM, and CB (middle)] compared to the LPS + PROT group (Figure S1 in Supplementary Material), indicating more animals with increased heterogeneity in their diffusivities within the ROIs. Given the greater variability in injury in the LPS + INJ group, we applied a colour map threshold technique to better distinguish between the two ventilation groups.

Colour map techniques were used to describe the tensor orientation in relation to the white matter tract direction (29). Histograms of AD and RD for the whole brains of individual lambs in the two groups exposed to LPS and two groups without LPS treatment (6) are plotted in Figure 2. The threshold range that highlighted lower diffusivity voxels (Figure 2, green dashed line) was manually set below the 90th percentile of intensities in the protective lambs and hence corresponded to the red shading in the injured lambs as shown in the examples of Figures 6 and 7.

In the low diffusivity maps obtained from the lambs exposed to LPS during the perinatal period, the red shading (threshold range) was more visible in the Th of AD and RD maps (Figure S2 in Supplementary Material) in three LPS + INJ lambs, while the lowest diffusivity values were almost absent in the same region in the LPS + PROT group. Further, there were low AD and RD values in the amygdala of one LPS + INJ lamb (Figure S2 in Supplementary Material). In the FWM, there were three LPS + INJ lamb brains that showed lower AD values compared to those in the LPS + PROT group (Figure S3 in Supplementary Material). Further, low RD values appeared in the FWM of two LPS + INJ lambs compared to the LPS + PROT group lambs (Figure S3 in Supplementary Material). We also found exceptionally low AD and RD voxel values, as shown in red and black (below the threshold range), in the IC, putamen, caudate, and septum of lambs in the LPS + INJ group, but rarely in the LPS + PROT group (Figures 6 and 7). In the PVWM, low AD and RD voxel diffusivities were found in the LPS + INJ group when compared to LPS + PROT group lambs in addition to grey matter regions of the cerebellum, which were lower in diffusivity than white matter (Figure S4 in Supplementary Material). In all ROIs, the calculated MD maps had similar patterns of distribution of low diffusivity as shown in the AD and RD maps (Figures S4 and S5 in Supplementary Material). Therefore, the colour maps revealed that the low-intensity voxels for AD, RD, and MD closely correlated with white matter regions in most of the cerebral anatomical structures that were chosen.

As a control reference, we applied the colour map technique to lambs that underwent the same ventilator and MRI protocols but were not exposed to LPS (PROT and INJ control lambs); the physiology and other MRI data have been reported previously (6). Lambs from our previous study did not differ in body weight, blood gas parameters, or HR at the time of intubation and cord clamping (time 0; Table 1) with the lambs from this study. We found no low AD values in the FWM, IC, putamen, caudate, septum, and PVWM for either PROT or INJ groups, while low RD values appeared in some regions of white matter of INJ lambs when compared to PROT group (Figures S6 and S7 in Supplementary Material). Finally, quantifying the number of voxels found to have low diffusivity within each ROI, we compared counts between LPS and control lambs (Figure 8). There were frequently lower AD, RD, and MD intensities in the Th, IC, PVWM, and FWM in a subgroup of LPS + INJ lambs compared to LPS + PROT, as well as when compared to controls when compared to controls that were not exposed to LPS.